I build web applications end to end — REST APIs, database schemas, authentication, and the frontend that sits on top of it. Most of my production work is Node.js/Express/MongoDB/React, but I'm not tied to the stack — I also work in Python, mostly for backend tooling and, more recently, computer vision, and I'm currently working on a Next.js/Node.js production stack during my internship.
I'm currently a software engineering student, splitting time between coursework, an internship focused on infrastructure and deployment for a production business application, and independent projects where I get to make my own architectural calls. The independent projects tend to be where I push furthest technically — that's where a marketplace app with real role-based auth turned into a Python video pipeline with its own face-detection and clustering logic.
|
🎬 FluxCutter A desktop tool that finds a specific person across a video and cuts their appearances into one reel. The full pipeline — detect → track → group → export — runs end to end via CLI and a desktop UI. Current focus is keyframe-based seeking, to cut down on the decode cost that dominates runtime today. |
⚙️ Deployment Infrastructure Evaluating and documenting self-hosted PaaS setups (Coolify on Hetzner Cloud) for internship deployment work — comparing a single-VPS setup against a hybrid local-machine/remote-Hetzner option, and writing up the architecture decisions as I go. |
Languages
Frontend
Backend
Databases
AI / ML & Computer Vision
Tools & Platforms
🎬 FluxCutter
A desktop application that isolates one person's appearances across a video and compiles them into a single exported reel. It detects faces (OpenCV YuNet), links detections into tracks, groups tracks into identities using ArcFace embeddings, and exports a merged clip through an in-process PyAV pipeline.
What makes it worth pointing to isn't just that it works — it's the engineering behind getting it to work reliably: memory profiling that cut peak usage from ~5GB to ~1GB, a documented RGB/BGR channel-swap bug that was silently dropping detections, and 115 tests covering the pipeline without needing a display.
🛒 Nile Market
A full-stack campus marketplace connecting student buyers and sellers, deployed live. Built with proper production concerns in mind: JWT auth, role-based access control, rate limiting, Helmet for secure headers, and listing moderation before anything goes public.
🥗 NutritionOS
A MERN-based meal planning and health tracking platform — BMI/calorie monitoring with personalized meal recommendations, secure authentication, and progress tracking on a responsive dashboard.
🎥 Recommendation Agent
A content-based movie recommendation system using similarity scoring over user preferences, built in a Jupyter notebook and deployed as an interactive app via Streamlit.
Outside of my own repos, I've contributed pull requests to other developers' projects that have been merged in (reflected in GitHub's Pull Shark achievement).
- Keyframe-based video seeking, to replace brute-force frame decoding in FluxCutter
- Backend architecture and system design at a deeper level
- Self-hosted deployment infrastructure — Coolify on Hetzner Cloud